Capacity Forecast Report Fall Update Sean Chang Market Analysis and Design Suresh Pabbisetty CQF, ERP, CSQA Credit CWG/MCWG December 16, 2015 ERCOT Public

2 Capacity Forecast Model - History ERCOT Public ERCOT began posting of the Capacity Forecast Model results starting from February 24 tth The report shows: Hourly forecast MW excess reserves for OD+1 through OD+6, with partial day forecast for OD+7. For OD-1 through OD-7, Hourly MW forecast for the Operating Day generated on the OD-1 relative to that day (the most recent capacity forecast for that day) Actual MW excess reserves for that hour HUBBUSAVG price for the hour Posted forecast values are P 50.

3 The idea is that Reserve Forecast could help to identify potential scarcity events during the forward week. With the implementation of the Operating Reserve Demand Curve (ORDC), we expect an increased correlation between reserves and prices. Reserves forecast calculation: Expected Reserves = + Non-Wind Resource HSL + AWS Wind Forecast + Load Resource RRS – Resource Outages – Load Forecast ± Load Forecast Error Capacity Forecast Model – Model recap

4 Model Run: Capacity Forecast Model was run at 8 AM for Operating Days March 1, 2015 through November 30, 2015 with P 50, and P 90 exceedance confidence intervals. Definitions: Exceedance Confidence Interval, P xx is xx% confidence that actual excess reserves will exceed modeled excess reserves Model Error = Modeled estimated reserves - actual reserves OFF n indicates forecast model run “n” days in advance of Operating Date. Interpretation: Positive model error indicates failure to predict while negative model error indicates false positive. Capacity Forecast Model – Model recap

5 Correlation of Capacity Forecast Reserves to Actual Reserves ERCOT Public

6 OFF4 P50 vs. Actual Reserves Correlation (MAR 2015-NOV 2015)

7 OFF1 P50 vs. Actual Reserves Correlation (MAR 2015-NOV 2015)

8 OFF4 P50 vs. Actual Reserves Correlation (Summer 2015)

9 OFF1 P50 vs. Actual Reserves Correlation (Summer 2015)

10 OFF4 P50 vs. Actual Reserves Correlation (Fall 2015)

11 OFF1 P50 vs. Actual Reserves Correlation (Fall 2015)

12 Capacity Forecast Model Excess Reserves vs Actual Results Correlation Summary: An Exceedance Confidence Interval of P 50 Same Exceedance Confidence Interval is used for the corresponding reports published on MIS since late February Capacity Forecast Model – Correlation Summary PeriodOFF4 CorrelationOFF1 Correlation Mar 2015-Nov %92% Spring %83% Summer %96% August %96% Fall %92%

13 Capacity Forecast Model Reserves Error Statistics ERCOT Public

14 Error = Forecast Reserves – Actual Reserves Error < 0 means Model predicted to have reserves lower than what they actually turned out to be Type I Error or False Positive Potential to ask for additional collateral but risk has not materialized Error > 0 means Model predicted to have reserves higher than what they actually turned out to be Type II Error or False Negative Potential failure to ask for additional collateral when risk has materialized Note: Various combinations of Exceedance Confidence Interval (P50 vs P90), offsets (OFF1 vs OFF4) and Excess Reserves Thresholds (ERT MW) could be adjusted to model for desired level of risk. Capacity Forecast Model – Error

15 Capacity Forecast Model OFF4 Reserves Error Statistics (in MW) Mar 2015-Nov 2015 Spring 2015Summer 2015Aug 2015Fall 2015 Mean (P50) (139) (1,503) Mean (P90) (2,910) (2,391) (1,980) (2,225) (4,374) Std. Dev. (P50) 3,468 4,148 2,600 2,352 3,032 Std. Dev. (P90) 3,562 4,179 2,752 2,516 3,122 Min (P50) (13,681) (9,350) (8,273) (12,217) Min (P90) (16,414) (16,035) (13,219) (12,142) (16,414) Max (P50) 13,946 9,884 6,591 7,589 Max (P90) 10,846 7,524 4,090 5,218

16 Capacity Forecast Model OFF1 Reserves Error Statistics (in MW) Mar 2015-Nov 2015 Spring 2015Summer 2015Aug 2015Fall 2015 Mean (P50) (1,057) (746) (881) (1,307) (1,550) Mean (P90) (2,962) (2,528) (2,949) (3,376) (3,414) Std. Dev. (P50) 2,479 2,715 2,199 2,279 2,420 Std. Dev. (P90) 2,597 2,802 2,384 2,464 2,513 Min (P50) (13,054) (9,727) (8,731) Min (P90) (14,946) (12,717) (11,384) Max (P50) 9,613 6,455 4,534 7,366 Max (P90) 8,026 4,845 3,241 5,850

17 Error comparison across Exceedance Confidence Intervals: Mean error with P90 is higher than that of P50 Standard Deviation of error with P90 and P50 are close Error comparison across seasons: Standard Deviation of error is relatively higher for Spring 2015 than that of Summer 2015 with Fall 2015 in the middle Error comparison across OFFset periods: Standard Deviation of error is relatively lower for OFF1 than that of OFF4 Note: Please note that these comparisons are only based on a very limited dataset of model run. A larger dataset may be required for statistical significance. Capacity Forecast Model – Error Statistics comparison

18 Capacity Forecast Model Reserves Error Contributing Factors ERCOT Public

19 Primary Factors contributing to Model Error: Net Load Error: Net of Load Forecasting Error and Wind Forecasting Error Some negative correlation is observed between Load Forecast Error and Wind Forecast Outage Error: Units recorded in Outage Scheduler as having outages are actually online during the same period HSL Error: Telemetered HSL of units is different than the forecast HSL Capacity Forecast Model – Error Contributing Factors

20

21

22

23

24

25

26 Capacity Forecast Model Reserves vs RTSPP Relation ERCOT Public

27 log Forecasted Reserves to log RTSPP Correlation: Capacity Forecast Model – Reserves to RTSPP Correlation PeriodOFF#P xx R^2 Mar Nov OFF1 P5042.0% P9042.5% OFF4 P5035.0% P9036.0% Spring 2015 OFF1 P5018.4% P9018.6% OFF4 P509.4% P909.6% Summer 2015 OFF1 P5068.4% P9069.8% OFF4 P5067.2% P9068.4% Fall 2015 OFF1 P5034.1% P9035.1% OFF4 P5028.7% P9030.6%

28 Capacity Forecast Model – Reserves to RTSPP Correlation

29 Capacity Forecast Model – Reserves to RTSPP Correlation

30 Capacity Forecast Model – Reserves to RTSPP Correlation

31 Capacity Forecast Model – Reserves to RTSPP Correlation

32 Capacity Forecast Model – Reserves to RTSPP Correlation

33 Capacity Forecast Model – Reserves to RTSPP Correlation

34 Capacity Forecast Model – Reserves to RTSPP Correlation

35 Capacity Forecast Model – Reserves to RTSPP Correlation

36 Capacity Forecast Model Results Summary ERCOT Public

37 Model predictability is relatively better in Summer than Spring A summary of effects of various model Inputs is as follows; Capacity Forecast Model – Results Summary Exceedance Confidence Interval Model AttributeP50P90 Probability of False PositivesRelatively LowerRelatively Higher Probability of False NegativesRelatively HigherRelatively Lower OFFset Model AttributeOFF4OFF1 Probability of False PositivesRelatively HigherRelatively Lower Probability of False NegativesRelatively HigherRelatively Lower Threshold Level Model Attribute3000MW5000MW Probability of False PositivesRelatively LowerRelatively Higher Probability of False NegativesRelatively HigherRelatively Lower

38 Correlation of Forecast Reserves to Actual Reserves is really strong in Summer and it is relatively weak in Spring Correlation of log Forecast Reserves to log HUBBUSAVG RTSPP is around 70% in the summer Capacity Forecast Model could be one of the potential options for reasonably predicting price risk Capacity Forecast Model – Results Summary continued.. Note: Supplemental graphs/charts available in APPENDIX.

39 Questions

40 Appendix Additional Graphs/Charts

41 Correlation of Capacity Forecast Reserves to Actual Reserves ERCOT Public

42 OFF4 P50 vs. Actual Reserves Correlation (Spring 2015)

43 OFF1 P50 vs. Actual Reserves Correlation (Spring 2015)

44 OFF4 P50 vs. Actual Reserves Correlation (August 2015)

45 OFF1 P50 vs. Actual Reserves Correlation (August 2015)